You are on page 1of 5

week ending

PRL 112, 145502 (2014) PHYSICAL REVIEW LETTERS 11 APRIL 2014

Finite Size Analysis of Zero-Temperature Jamming Transition under Applied Shear


Stress by Minimizing a Thermodynamic-Like Potential
Hao Liu, Xiaoyi Xie, and Ning Xu*
CAS Key Laboratory of Soft Matter Chemistry, Hefei National Laboratory for Physical Sciences at the Microscale,
and Department of Physics, University of Science and Technology of China, Hefei 230026, Peoples Republic of China
(Received 9 December 2013; published 10 April 2014)
By finding local minima of a thermodynamic-like potential, we generate jammed packings of frictionless
spheres under constant shear stress and obtain the yield stress y by sampling the potential energy
landscape. For three-dimensional systems with harmonic repulsion, y satisfies the finite size scaling with
the limiting scaling relation y c; , where c; is the critical volume fraction of the jamming
transition at 0 in the thermodynamic limit. The finite size scaling implies a length c;
with 0.81  0.05, which turns out to be a robust and universal length scale exhibited as well in the
finite size scaling of multiple quantities measured without shear and independent of particle interaction.
Moreover, comparison between our new approach and quasistatic shear reveals that quasistatic shear tends
to explore low-energy states.

DOI: 10.1103/PhysRevLett.112.145502 PACS numbers: 61.43.-j

At zero temperature and shear stress, a packing of that can sustain the applied shear stress. Or in practice, the
frictionless spheres interacting via repulsions jams into a probability of finding such jammed states is low. If states
disordered solid when its volume fraction exceeds constrained at the desired shear stress were quickly
a critical value c at the so-called point J [13]. As a generated, we would be able to sample the potential energy
simplified model to understand the noncrystalline liquid- landscape and locate the jamming transition at > 0, in a
solid transition of various materials including granular similar way to what was done for the jamming transition at
materials, foams, colloids, emulsions, and glasses, jammed 0 [3]. However, such an approach is apparently
packings of frictionless spheres exhibit interesting but lacking.
unusual critical behaviors at point J [317]. In this Letter, we report that the sampling of the potential
In addition to the volume fraction, shear stress and energy landscape under constant shear stress can be
temperature T are the other two control parameters to cause realized by minimizing a thermodynamic-like potential.
a generalized jamming transition, i.e., yielding and glass The yield stress determined from the probability of finding
transition [1]. A jammed solid remains rigid when subject jammed states is scaled well with the volume fraction. The
to a shear stress smaller than the yield stress y , while it finite size scaling of the yield stress indicates that point J is
unjams and flows otherwise. It has been shown that the a critical point associated with a diverging length. We find
yield stress of the T 0 jammed solids decreases when the same length scale in the finite size scaling of multiple
decreasing the volume fraction and vanishes at point J quantities under zero shear stress and with different particle
[7,12,18,19]. This is different from the glass transition interactions, implying the universality of the length scale.
temperature through the fact that in the T 0 limit the Moreover, by comparing properties of jammed states
glass transition occurs at a volume fraction lower than c obtained from our new approach and quasistatic shear,
[2026]. Therefore, point J is more relevant to the volume we find that quasistatic shear explores low-energy states in
fraction and shear stress than to the temperature. Jammed the potential energy landscape, which may provide us with
packings of frictionless spheres under applied shear stress a possible way to search for ultrastable glasses.
thus serve as typical systems to study the criticality of point Our systems are three dimensional with side length L
J [7,8,12]. (from L=2 to L=2) in all directions. Lees-Edwards
In most of the previous simulations, the yield stress of a boundary conditions are applied to mimic shearing [29],
jammed solid has been defined as either the average shear with the shear force and shear gradient in the x and y
stress of the quasistatic shear flow in which the shear stress directions, respectively. The system contains N=2 large and
is not a controllable parameter [18,19,27,28] or the critical N=2 small spheres with equal mass m and a diameter ratio
shear stress extrapolated from nonequilibrium molecular 1.4 to avoid crystallization. The interaction potential
dynamics simulations above which the system loses shear between particles i and j is
rigidity and flows forever [18,28]. In the potential energy    
rij rij
landscape perspective, the yield stress corresponds to the U ij 1 1 ; (1)
critical shear stress above which there is no jammed state dij dij

0031-9007=14=112(14)=145502(5) 145502-1 2014 American Physical Society


week ending
PRL 112, 145502 (2014) PHYSICAL REVIEW LETTERS 11 APRIL 2014
1
where rij is their separation, dij is the sum of their radii, and (a) (b)
x is the Heaviside function. Here we only show results 0.8
= 0.635
for harmonic repulsion with 2. To obtain jammed 0.6
= 0.640
= 0.650
states at the desired shear stress , we start with random = 0.660

f
0.4 = 0.680
states and minimize a thermodynamic-like potential for = 0.700
sheared nonequilibrium systems: 0.2

0
H~r1 ; ; ~rN ; U~r1 ; ; ~rN ; L3 (2) 0 0.001 0.002 0.003 0.004 0 0.0005 0.001 0.0015

using the fast inertial relaxation


P PN engine minimization FIG. 1 (color online). (a) Probability of finding jammed states, f,
method [30], where U N1 i1 ji1 U ij is the internal measured as a function of applied shear stress with the cutoff strain
energy, is the shear strain, and ~ri is the location of particle c 1 for N 64 systems. The solid curves are the fits using
i. We set the units of mass, energy, and length to be m, , Eq. (3). (b) Probability measured at 0.650 with c 1, 2, and 4
(solid curves from the left to the right). The dashed and dot-dashed
and small particle diameter ds.
curves are predicted c 2 and c 4 curves using Eq. (4).
Equation (2) is a Legendre transformation to change the
independent variable from the shear strain to the shear
stress. When a shear stress is applied, the shear forces and does not have any difference from that under a strain of ,
displacements at y L=2 are F ~  L2 x and where l is an integer. In this sense, c 1 is a natural and
~s L=2x. The energy caused by the shear stress is reasonable choice. Of course, y and w vary with c .
~ ~s L3 , which leads to the
~ ~s F However, this variation is trivial and can be predicted
thus equal to F
simply from the recurrence relation
second term on the right-hand side of Eq. (2). Both the
shear stress and pressure p can be controlled simultane- f l1 f l f1 f l ; (4)
ously if a third term pL3 is added to the right-hand side of
Eq. (2), which should have significant applications to assuming that from l to l 1 those 1 f l
systems under both constant shear stress and constant unjammed states have another chance to go to jammed
pressure. Here, we are concerned about how properties states with a probability of f, where f, f l , and f l1 are the
of packings of frictionless spheres vary with the volume probabilities of finding jammed states at the same shear
fraction, so the volume fraction or length L is the control stress with c 1, l, and l 1. In Fig. 1(b), we compare f l
parameter instead of the pressure. obtained from the direct measure by setting c l and that
The shear strain is initially set to be zero. When the predicted from Eq. (4). They agree well. Therefore, the
applied shear stress is much smaller than the yield stress, probability of finding jammed states measured within a unit
our algorithm can quickly find jammed states at small . strain, f, reflects nontrivially the fraction of configurational
When increasing the shear stress, a larger is needed. space occupied by jammed states under the constraint of
When > y , because there is no static state sustainable to constant shear stress.
the shear stress, goes to infinity. It is impracticable to run Figure 2(a) shows the volume fraction dependence of the
simulations up to extremely large . We thus set a cutoff yield stress for different system sizes. In the region from
strain c . If the simulation does not find any local energy 0.65 to 0.70, the yield stress is linearly scaled with the
minima before c , the initial random configuration corre- volume fraction
sponds to an unjammed state. Otherwise, it leads to a
jammed state. For each pair of and , we enumerate the y ; N 0;N c;N ; (5)
number of jammed states over 10 000 independent runs,
from which the probability of finding jammed states, where 0;N and c;N are fitting parameters. As shown in
f; , is determined. Fig. 2(b), 0;N and c;N can be well fitted with
Figure 1(a) shows f; with c 1 for N 64
systems. At fixed , we approximate f; into a c;N c; 0.137N 0.457 ; (6)
complementary error function
  0;N 0; 0.082N 0.363 ; (7)
1 y
f; erfc p ; (3)
2 2w where c; 0.649 and 0; 0.018 are the critical
volume fraction of point J and the slope of y in the
where y and w are the mean value and standard deviation, thermodynamic limit. The value of c; is consistent with
which are defined as the yield stress and its width. previous studies of the same bidisperse systems [11,25].
With Lees-Edwards boundary conditions, the shear For finite size systems, y ; N deviates from Eq. (5) at
strain has a period of 1. A state under a strain of l low volume fractions. As illustrated by Fig. 2(a), this

145502-2
week ending
PRL 112, 145502 (2014) PHYSICAL REVIEW LETTERS 11 APRIL 2014

10-1
(a) (b)
c, c,N
observation of the decay of the gap between two yield
10
-3 0,N 0, stresses approaching point J [18]. In the high volume
fraction
p regime studied here, w is inversely proportional to
y

10-2 N . Thepdecay of w at low volume fractions is even faster


10
-4
than 1= N . When increasing the system size, w; N
N = 64
N = 256
N = 1024
-5
N = 4096 shows the tendency to approach the form
10 10-3 1
0.62 0.64 0.66 0.68 0.7 10 102 103 104
1
-2
N w; N p c;N : (10)
10 N
100 (c) (d)

We thus claim that in the thermodynamic limit the yield


0.41

10-1
0.5
stress is well defined with w 0. The gap between two
yn N

wN
-3 yield stresses should be a finite size effect, as already
10
10-2 suggested in Ref. [28].
To check whether the length described by Eq. (9) is
0 0.5 1 10-3 10-2 10-1 universal or just specific to the yield stress, we perform the
0.41
( c,) N c,N finite size analysis of typical quantities mostly of concern in
the study of jamming under zero shear stress. For each pair
FIG. 2 (color online). (a) Volume fraction and system size N of and N, we generate 10 000 independent states without
dependence of the yield stress y . The solid curves are the fits to
shear at T 0 (including jammed and unjammed) and then
the high data using Eq. (5). (b) System size dependence of the
fitting parameters c;N and 0;N in Eq. (5). The dashed lines do the average. Figure 3 demonstrates that the potential
are the fits using Eqs. (6) and (7). (c) Finite size scaling of the energy per particle u U=N, pressure p, coordination
reduced yield stress yn . (d) Volume fraction number z, and shear modulus G all show very nice finite
pdependence of the
width of the yield stress w multiplied by N. The dot-dashed size scaling with the same length scale as proposed by
line shows the scaling of Eq. (10). Eq. (9):

u c; 2 gu c; N 0.4 ; (11)
deviation is weaker when increasing the system size, which
is likely associated with the system size dependence of the p
c distribution [3,31]. c; gp c; N 0.4 ; (12)
2
The critical scaling of Eq. (5) inspires us to perform finite
size scaling of the yield stress. As shown in Fig. 2(c), the z zc; c; 0.5 gz c; N 0.4 ; (13)
yield stress indeed exhibits excellent finite size scaling,
again suggesting that point J at c; is critical. All the yield
stress data collapse nicely onto a master curve in the
following form: 100 100
(a) (b)
10-1
y -1
yn c; g c; N  10
0.4

(8) 10-2
0;N
(p/ ) N
0.8
uN

10-3
2

N = 64 10-2
-4 N = 256
with the limiting scaling of the yield stress being 10
N = 1024
y c; , where 0.41  0.02 is obtained to best 10-5
N = 4096 10-3
collapse all the data. Equation (8) implies a length 10
diverging at c; in the form (c) (d)
10-1
0.2

c; ; (9) 0
(z - zc,)

0.2
GN

10-2
where 1=3 0.81  0.05. Later we will show that -10
this length is not limited to the yield stress. Finite size
scaling of multiple quantities measured without shear -20 10-3
0 0.5 1 0 0.5 1
exhibits the same length scale. ( c,) N
0.4
( c,) N
0.4

In previous studies [18,28], two yield stresses have been


observed from the liquid and solid perspectives, respec- FIG. 3 (color online). Finite size scaling of (a) potential energy
tively. The width of the yield stress w directly reflects this per particle u, (b) pressure p, (c) excess coordination number
uncertainty. Figure 2(d) shows that w decreases when beyond isostaticity z zc;, and (d) shear modulus G for systems
decreasing the volume fraction, consistent with the without shear.

145502-3
week ending
PRL 112, 145502 (2014) PHYSICAL REVIEW LETTERS 11 APRIL 2014

G c; 0.5 gG c; N 0.4 ; (14) As a comparison, we also show the results for quasistatic
shear sampling in Fig. 4. To mimic quasistatic shear, we
where zc; 6 is the isostatic value. We tune the exponent successively deform jammed states from 0 to 1 using a
of N from 0.41 to 0.4 to have the best data collapse. The step strain 104 followed by the potential energy
limiting scaling relations of these quantities are well known minimization. Ten thousand jammed states with different
for marginally jammed solids with harmonic repulsion shear stresses are obtained during one course of quasistatic
[3,13,15,25]. Equations (11) and (12) are simply related via shear, from which the shear stress dependence of the
p 2 dU=d, which leads to the factor 2 in Eq. (12) potential energy and coordination number can be achieved.
and the relation gp x 2gu x xgu0 x. The same length The results in Fig. 4 are from 1000 independent runs of
scale from the finite size scaling of the yield stress, quasistatic shear. In contrast to our random sampling,
potential energy, pressure, coordination number, and shear quasistatic shear leads to a decrease of both the potential
modulus suggests that the scaling exponent of the length energy and coordination number when increasing the shear
found here is universal for three-dimensional jammed states stress at low shear stresses. At all shear stresses, jammed
with harmonic repulsion. states found by quasistatic shear sampling always have
It has been shown that most of the scaling relations of lower potential energy and coordination number than those
marginally jammed states depend on the exponent in obtained from random sampling. This discrepancy implies
Eq. (1) of the particle interaction, except for the co- the biased nature of the quasistatic shear to sample the
ordination number [3]. It is thus possible that the finite potential energy landscape: it tends to explore low-energy
size scaling for the coordination number shown in Fig. 3(c) states.
also works for other interparticle potentials like Hertzian The bias of quasistatic shear sampling contains some
repulsion [ 2.5 in Eq. (1)]. If then, the same length scale interesting implications. For jammed states interacting via
would be observed in the finite size scaling of multiple repulsions, lower coordination number and potential energy
quantities for Hertzian repulsion as well. We repeat Fig. 3 mean that the states are closer to the unjamming transition
for Hertzian repulsion (see the Supplemental Material [32]) subject to the change of volume fraction. This explains why
and indeed find the same length scale. Therefore, the length the critical volume fraction of the jamming transition
scale reported here may be independent of interparticle determined from quasistatic shear sampling is higher than
potential, at least for harmonic and Hertzian repulsions. that from random sampling at 0 [7,19].
Minimizing the thermodynamic-like potential enables us More interestingly, for the case shown in Fig. 4, quasi-
to sample jammed states under the desired shear stress. static shear can find states with potential energy about 35%
Because the initial random states before minimization are lower than the normal value. This is actually analogous to
independently selected, our sampling of the potential the fact that inherent structures with lower potential energy
energy landscape is unbiased. We can thus have an can be explored by supercooled liquids with slower cooling
unbiased statistical picture of how the properties of jammed rate [33]. However, the advantage of quasistatic shear is
solids vary with shear stress. In Fig. 4, we show the that it can overcome energy barriers easily and thus
potential energy per particle and coordination number efficiently speed up the search of low-energy states.
averaged over jammed states under the same shear stress Recently, it has been reported that ultrastable glasses with
obtained from the unbiased random sampling. When low energy and aged over thousands of years can be
increasing the shear stress, both the potential energy and quickly obtained from a vapor deposition method
coordination number show a plateau at low shear stresses [34,35]. Since quasistatic shear tends to explore low-energy
and shoot up near yielding. states, it may provide us with an alternate efficient way to
search for ultrastable glasses.
(a) (b)
In conclusion, we sample the potential energy landscape
0.06 6.9 under the constraint of constant shear stress by minimizing
a thermodynamic-like potential. Using this new approach,
6.8 we obtain the yield stress of jammed solids from measuring
z
u

0.04
the probability of finding jammed states under constant
6.7
shear stress. The yield stress and multiple quantities
0.02 measured without shear all show very nice finite size
0 0.0002 0.0004 0 0.0002 0.0004 scaling, from which we obtain a universal length scale

described in Eq. (8). Multiple length scales have been
FIG. 4 (color online). Shear stress dependence of (a) the reported for the jamming transition at point J in different
potential energy per particle u and (b) coordination number z measurements [311], which may also be one of the most
obtained from random sampling (circles) and quasistatic shear special and elusive features of the criticality of point J. The
sampling (squares). The systems consist of N 1024 particles at length scale reported here is robust for three-dimensional
0.66. The solid curves are to guide the eye. systems because it is associated with multiple quantities

145502-4
week ending
PRL 112, 145502 (2014) PHYSICAL REVIEW LETTERS 11 APRIL 2014

and is possibly independent of the interparticle potential. [14] N. Xu, V. Vitelli, M. Wyart, A. J. Liu, and S. R. Nagel, Phys.
Moreover, we propose to look for low-energy ultrastable Rev. Lett. 102, 038001 (2009); V. Vitelli, N. Xu, M. Wyart,
glasses using quasistatic shear because it can explore low- A. J. Liu, and S. R. Nagel, Phys. Rev. E 81, 021301 (2010).
energy states efficiently. [15] C. Zhao, K. Tian, and N. Xu, Phys. Rev. Lett. 106, 125503
(2011).
We are grateful to Kunimasa Miyazaki and Stephen Teitel [16] A. S. Keys, A. R. Abate, S. C. Glotzer, and D. J. Durian,
for helpful discussions. This work is supported by National Nat. Phys. 3, 260 (2007).
Natural Science Foundation of China Grants No. 21325418 [17] D. A. Head, Phys. Rev. Lett. 102, 138001 (2009).
and No. 11074228, National Basic Research Program of [18] M. Pica Ciamarra and A. Coniglio, Phys. Rev. Lett. 103,
235701 (2009).
China (973 Program) No. 2012CB821500, CAS 100- [19] C. Heussinger and J.-L. Barrat, Phys. Rev. Lett. 102, 218303
Talent Program No. 2030020004, and Fundamental (2009).
Research Funds for the Central Universities Grant [20] A. Ikeda, L. Berthier, and P. Sollich, Phys. Rev. Lett. 109,
No. 2340000034. 018301 (2012).
[21] F. Krzakala and J. Kurchan, Phys. Rev. E 76, 021122
(2007).
[22] L. Berthier and T. A. Witten, Phys. Rev. E 80, 021502
*
ningxu@ustc.edu.cn (2009); Europhys. Lett. 86, 10001 (2009).
[23] G. Parisi and F. Zamponi, Rev. Mod. Phys. 82, 789 (2010).
Present address: Department of Physics, New York Uni-
versity, New York, NY 10012, USA [24] Z. Zhang, N. Xu, D. T. N. Chen, P. Yunker, A. M. Alsayed,
[1] A. J. Liu and S. R. Nagel, Nature (London) 396, 21 (1998). K. B. Aptowicz, P. Habdas, A. J. Liu, S. R. Nagel, and A. G.
[2] M. van Hecke, J. Phys. Condens. Matter 22, 033101 (2010). Yodh, Nature (London) 459, 230 (2009).
[3] C. S. OHern, L. E. Silbert, A. J. Liu, and S. R. Nagel, Phys. [25] L. Wang and N. Xu, Soft Matter 9, 2475 (2013).
Rev. E 68, 011306 (2003). [26] P. Olsson and S. Teitel, Phys. Rev. E 88, 010301(R)
[4] L. E. Silbert, A. J. Liu, and S. R. Nagel, Phys. Rev. Lett. 95, (2013).
098301 (2005). [27] L. Berthier and J.-L. Barrat, J. Chem. Phys. 116, 6228
[5] M. Wyart, L. E. Silbert, S. R. Nagel, and T. A. Witten, Phys. (2002).
Rev. E 72, 051306 (2005); M. Wyart, S. R. Nagel, and T. A. [28] N. Xu and C. S. OHern, Phys. Rev. E 73, 061303 (2006).
Witten, Europhys. Lett. 72, 486 (2005). [29] M. P. Allen and D. J. Tildesley, Computer Simulation of
[6] W. G. Ellenbroek, E. Somfai, M. van Hecke, and W. van Liquids (Oxford University Press, New York, 1987).
Saarloos, Phys. Rev. Lett. 97, 258001 (2006). [30] E. Bitzek, P. Koskinen, F. Ghler, M. Moseler, and P.
[7] P. Olsson and S. Teitel, Phys. Rev. Lett. 99, 178001 (2007); Gumbsch, Phys. Rev. Lett. 97, 170201 (2006).
D. Vgberg, D. Valdez-Balderas, M. A. Moore, P. Olsson, [31] N. Xu, J. Blawzdziewicz, and C. S. OHern, Phys. Rev. E
and S. Teitel, Phys. Rev. E 83, 030303(R) (2011). 71, 061306 (2005).
[8] T. Hatano, J. Phys. Soc. Jpn. 77, 123002 (2008). [32] See Supplemental Material at http://link.aps.org/
[9] C. P. Goodrich, A. J. Liu, and S. R. Nagel, Phys. Rev. Lett. supplemental/10.1103/PhysRevLett.112.145502 for finite
109, 095704 (2012). size scaling of systems with Hertzian repulsion.
[10] J. A. Drocco, M. B. Hastings, C. J. Olson Reichhardt, and C. [33] S. Sastry, P. G. Debenedetti, and F. H. Stillinger, Nature
Reichhardt, Phys. Rev. Lett. 95, 088001 (2005). (London) 393, 554 (1998).
[11] M. Ozawa, T. Kuroiwa, A. Ikeda, and K. Miyazaki, Phys. [34] S. F. Swallen, K. L. Kearns, M. K. Mapes, Y. S. Kim, R. J.
Rev. Lett. 109, 205701 (2012). McMahon, M. D. Ediger, T. Wu, L. Yu, and S. Satija,
[12] B. P. Tighe, E. Woldhuis, J. J. C. Remmers, W. van Saarloos, Science 315, 353 (2007).
and M. van Hecke, Phys. Rev. Lett. 105, 088303 (2010). [35] S. Singh, M. D. Ediger, and J. J. de Pablo, Nat. Mater. 12,
[13] D. J. Durian, Phys. Rev. Lett. 75, 4780 (1995). 139 (2013).

145502-5

You might also like